Skip to content

Annotations Source

🤖 AI Drafted (Not reviewed)

Iterate a document’s user highlights / notes / regions and emit one cropped input per annotation. Use upstream of any vision or LLM tool to make the AI operate on a marked region instead of the whole document. (#914)

Tool id annotations_source
Category source
Uses a language model no
Needs a generative model no
Runs over many items no
Item handling batch
Structured output no
Human-verified not yet

What it reads

This tool takes no inputs.

What it emits

Port Type Required What it is
Files (files) files One cropped file per annotation (PNG for image / PDF regions, .txt for text highlights). The crops live in a tmp dir and survive for the workflow run.
Documents (documents) json One document entry per crop: the original document metadata + annotation_id + crop_kind so downstream tools can link results back to the right annotation.
Count (count) number Number of annotation crops emitted.

Options

Option Type Default What it does
kinds array [“highlight”, “bookmark”] Annotation kinds to iterate. Defaults to highlight + bookmark (the spatial / span-anchored kinds); exclude ‘rating’ which carries no crop content.
min_rating number 0 Only emit annotations with rating >= this value. 0 (default) means include all. Use to focus the AI on the user’s most-important highlights.

The prompt it sends

This tool does not send a prompt — it does its work in code.